RNN-based optimal consensus of high-order heterogeneous nonlinear MAS with input constraints

Most existing works on optimality in the consensus of multi-agent systems (MAS) can only achieve inverse optimality, which may not be desirable for industrial applications. Meanwhile, the distributed optimal consensus of high-order heterogeneous nonlinear MAS with input constraints is a challenging problem due to nonlinearity, heterogeneity, and input constraints. To address the problem, we propose a recurrent neural network (RNN) method. The problem is described by receding-horizon optimization and reduced to a resolvable problem via the use of Taylor expansion for state prediction. Then, an RNN is developed, by which a dynamic distributed consensus protocol emerges. The corresponding theoretical guarantee is provided, and numerical experiments validate the efficacy of our method. The experimental comparison with existing works corroborates the advantages of our method.